Clinical Prediction Rules: Challenges, Barriers, and Promise

Logistics experts, telecommunications experts, and transportation planners all talk about the "last mile problem". In medicine, we have our own last mile problem. Promising clinical prediction rules (CPRs) to improve diagnosis and prognosis are developed, but never properly validated (apparently developing something of your own is more fun that validating someone else's tool). Accurate tools for diagnosis and prognosis are developed and validated, but never widely used, or "baked in" to the process of care. Some funding agencies (PCORI, I'm looking at you) actually forbid research on the validation of clinical prediction rules. Wallace and Johansen's excellent editorial hits the nail on the head: there is no shortage of proposed CPRs, but there is a shortage of validation studies, of integration into the primary care process, and of studies that evaluate the impact of their use. Take the example of community-acquired pneumonia. Multiple CPRs have been developed to facilitate diagnosis, but none other than the 1991 Heckerling score have been prospectively validated (and that was only validated in the original study). The CRB-65 score for prognosis is accurate, simple, and widely recommended by guidelines, but is unknown to US physicians, and has never even been validated in the US setting. It is not integrated into any medical records. And of course, electronic health record vendors are notoriously difficult to work with when it comes to integration of decision support. This editorial serves as a wake-up call to clinicians, researchers, EHR vendors and funding agencies that this is important work and deserves our attention.

Competing interests: I have written a book about clinical prediction rules, and edit a medical reference (Essential Evidence Plus) that includes about 500 that I have programmed.